Voice Keyboard

byPsychopath · Tools

Verified safeOpen sourceExclusive

Download APK

—

No reviews reviews

downloads

7.0

android

AI summary

Open-source voice keyboard that sends speech to any OpenAI-compatible Whisper API for transcription, with optional LLM post-processing to fix errors, shorten text, add emoji, or translate. Version 1.9.0 improves punctuation spacing, prevents recording loss when switching keyboards, adds in-keyboard mic permission requests, and reduces upload size by 3×.

Generated by AI. May contain inaccuracies.

About this app

Android keyboard (IME) for speech-to-text. Sends audio to any OpenAI-compatible Whisper endpoint of your choice — by default Groq (free tier available), can be pointed at OpenAI or any other compatible provider. Optional LLM post-processing.

Features

Voice Input

- Real-time voice recording with amplitude visualization - Processing queue — start a new recording immediately, previous ones transcribe in the background - Works with any OpenAI-compatible Whisper API; ships with Groq as the default endpoint (whisper-large-v3-turbo) - Configurable API endpoint, model, and language - Auto-start recording when keyboard opens - Custom vocabulary — add names and technical terms to bias recognition; helps with contacts and rare words - Drop period for single-word output — handy for voice search where a trailing period gets in the way

Post-Processing

- Fix errors — corrects punctuation, spelling, removes filler words (um, uh) - Shorten — makes text concise while keeping key points - Emoji — adds relevant emoji to your messages - Rhyme — rewrites dictated text as poetry - Translate — translates to any of the supported languages - Supports OpenAI and Claude as processing providers - Customizable prompts and temperature for each mode

Keyboard

- Send button (paper plane) — sends Ctrl+Enter for quick message sending in messengers - Accelerating backspace — hold to delete slowly at first, then faster - Clipboard bar stays visible after paste for repeated pasting - Graceful shutdown — if keyboard hides during recording, audio is finalized and transcribed to clipboard

General

- 17 interface and transcription languages - Light, Dark, and Auto themes - Long-press spacebar to switch keyboard - Built-in test recording in settings - App logs and crash reports - Auto-update from GitHub Releases

Setup

1. Install the APK 2. Go to Settings → System → Languages & input → On-screen keyboard 3. Enable "Voice Keyboard" 4. Open the app and enter a Whisper API key. The default endpoint is Groq — get a free key at console.groq.com/keys. You can also point the app at OpenAI's Whisper endpoint or any other OpenAI-compatible provider in the same screen. 5. (Optional) Configure post-processing with your OpenAI or Claude API key

Privacy

No analytics, no telemetry, no advertising. Audio is sent only to the transcription provider you configure, using your own API key.

License

MIT

keyboard speech-to-text voice input open source offline-ready productivity

What's new

v1.9.0Sep 16, 2026

Dictated text now handles spacing correctly, and the . ? ! keys remove the space before punctuation. - Recordings no longer get deleted while uploading when switching keyboards or themes. - Accidental double taps on the mic no longer produce unsendable recordings. - Failed recordings can be deleted from the keyboard: hold the resend key, then hold it again. - A busy microphone now shows a message instead of crashing the keyboard. - Uploads are three times smaller: faster results on mobile data, fewer timeouts. - Mic permission can now be requested right from the keyboard; after "Don't ask again" it opens the app's settings. - The settings screen shows whether the keyboard is enabled and offers buttons to enable or switch it. - Provider presets now fill in the address and a valid model. - Reasoning models work with post-processing, and temperature now applies to Claude too. - A recording made before an API key is entered is kept instead of being lost. - Wrong model names now show the provider's own message. - Empty post-processing results no longer insert a stray space or overwrite the clipboard. - Logs and crash reports can be shared from settings. - Translations completed in all 17 languages.